Automate tests against latest libLcm on nuget

[hahn-kev]

This PR does 2 things.

1. Let's us run the tests against a new build of libLcm from nuget
2. Runs automated tests when pushing changes and in PR branches

I'm hoping this will let us catch issues earlier than when a new FLEx release goes out. Right now this only tests against new libLcm changes, but we could potentially download nightly FLEx builds and test against those too.

after this PR is merged in then these tests would run on each change that comes after, you can see the workflow that runs here in my fork https://github.com/hahn-kev/flexlibs/actions/runs/30076896761/job/89429540536

---

AI Summary

Here’s the full picture of what we built.

Problem

flexlibs talks to FieldWorks through pythonnet, loading DLLs from an installed FLEx tree (registry → RootCodeDir). You wanted to:

1. Test against latest SIL NuGet packages without assuming only the install’s copies
2. Eventually run that on Windows CI (download/install FieldWorks, cache it)

Part 1 — Swappable dependencies (local)

Path overrides in FLExGlobals.py

Three env vars, checked before registry discovery:

Env var	Role
FLEXLIBS_FW_CODE_DIR	Folder with FieldWorks.exe / app DLLs
FLEXLIBS_FW_PROJECTS_DIR	Projects folder (required if code dir is set)
FLEXLIBS_ASSEMBLY_DIR	Optional DLL folder prepended on sys.path so those assemblies win

Default behavior (no env vars) is unchanged: still uses the Windows registry.

NuGet overlay

• tests/nuget-overlay/Overlay.csproj references latest prerelease SIL.LCModel*, SIL.Core*, SIL.WritingSystems
• restore.ps1 does dotnet publish → flat .fw-nuget-overlay/
• make test-nuget restores that overlay, sets FLEXLIBS_ASSEMBLY_DIR, runs pytest
• Extra args pass through, e.g. -k "not CustomFields"

Important constraint: NuGet only publishes LCM/SIL libs. App assemblies (FwUtils, FieldWorks.exe, ICU init) still come from a FieldWorks binary tree. Overlay = “latest libraries on top of a real FW install.”

Custom-fields test

FLEXLIBS_TEST_PROJECT / FLEXLIBS_TEST_CUSTOM_FIELD so you can point at your own project (needs an entry-level custom text field). CI skips that test.

---

Part 2 — Windows CI

Branch: ci/windows-fieldworks-tests
PR: hahn-kev#1

Workflow (.github/workflows/test-windows.yml)

On a windows-latest runner:

1. Python 3.12 + .NET SDK
2. Cache installer, install tree, sample project
3. Install-FieldWorks.ps1 — download offline 9.3.9.1 installer, silent /quiet /norestart ADDLOCAL=ALL, ensure registry
4. Discover real install path (turns out it’s C:\Program Files\FieldWorks9, not SIL\FieldWorks 9)
5. Fallback: GitHub BuildDir.zip if exe missing after install
6. Setup-SampleProject.ps1 — extract Sena 2 .fwbackup into projects dir
7. make test-nuget -k "not CustomFields" with env pointing at that FW tree

Result

• First green run: 4 passed, 1 deselected
• Later runs ~2 minutes with cache hit (vs ~7+ on cold install)

---

How the pieces fit

flowchart TD
  nuget[dotnet publish SIL packages] --> overlay[".fw-nuget-overlay"]
  installer[FW offline installer or cache] --> fwTree["FieldWorks9 + projects"]
  overlay --> envAsm["FLEXLIBS_ASSEMBLY_DIR"]
  fwTree --> envCode["FLEXLIBS_FW_CODE_DIR / PROJECTS_DIR"]
  envAsm --> flexlibs[flexlibs / pythonnet]
  envCode --> flexlibs
  flexlibs --> pytest[pytest smoke + project tests]

Loading

Locally: .\make.bat test-nuget -k "not CustomFields"
CI: same idea, but it provisions FieldWorks for you and caches it.